Research Fellow Department of Neurosurgery, Perelman School of Medicine, University of Pennsylvania Philadelphia, PA, US
Introduction: Adolescent idiopathic scoliosis (AIS) is the most common form of pediatric scoliosis, affecting 2-4% of adolescents and potentially leading to severe spinal deformities if untreated. Surgical outcomes after posterior spinal fusion (PSF) are highly variable, and predicting post-operative success is crucial for optimizing surgical strategies. This study aimed to develop and validate a machine learning model to predict post-operative thoracic Cobb angle (TCA) correction, focusing on identifying patients with a correction greater than 75%.
Methods: This project was funded by an NIH grant (R21AR075971). This retrospective study was supported by the STROBE, TRIPOD+AI, and CLAIM guidelines. Data were sourced from an institutional AIS registry consisting of 83 patients who underwent PSF between 2014 and 2022. Pre-operative and post-operative radiographic data, patient demographics, and surgical variables were collected. The machine learning model was trained using extreme gradient boosting (XGBoost) with hyperparameter tuning via grid search. Key variables included pre-operative TCA, lumbar Cobb angle (LCA), operative time, and estimated blood loss (EBL). Missing data, affecting less than 1% of the dataset, were imputed using random forest-based multiple imputation. Model performance was evaluated using accuracy, sensitivity, precision, F1-score, area-under-the-curve (AUC), and precision-recall curve.
Results: There were no significant differences in demographics and baseline characteristics (P > 0.05) between the two outcome groups. Of the 83 AIS patients, 43 (52%) achieved a post-operative TCA correction greater than 75%, while 40 patients (48%) had a correction of 75% or less. The XGBoost model achieved an overall accuracy of 94.1%, with an AUC of 0.99 and PRC of 0.99.
Conclusion : The machine learning model demonstrated excellent performance in predicting post-operative thoracic Cobb angle correction in AIS patients, achieving high accuracy, AUC, and PRC. This model may serve as a valuable tool for pre-operative planning, aiding surgeons and informing patients.